A Comprehensive Review of Learning-Based Anomaly Detection Techniques in IoT Security Systems

Abstract

     The Internet of Things (IoT) is increasingly integrated into critical systems such as healthcare, transportation, and smart cities, making it a prime target for cybersecurity threats. As traditional intrusion detection systems (IDS) struggle to handle the volume and diversity of IoT-generated data, machine learning (ML) and deep learning (DL) techniques have emerged as promising solutions. This paper presents a comprehensive review of recent ML and DL-based approaches for anomaly detection in IoT environments. It categorizes key techniques including convolutional neural networks (CNNs), long short-term memory (LSTM) networks, autoencoders, and hybrid models, examining their strengths, limitations, and suitability for various IoT domains. The review also highlights preprocessing techniques such as feature selection, principal component analysis (PCA), oversampling (e.g., SMOTE), and federated learning (FL), which are essential for handling imbalanced and distributed data. Furthermore, the paper discusses commonly used datasets, evaluation metrics, and emerging research challenges. This survey aims to provide researchers and practitioners with a structured overview of state-of-the-art techniques and guide the development of efficient, scalable, and secure IDS solutions for modern IoT networks.